feat(cdn): add metrics (#5181)

Author: sguiheux

engine/api/api.go

@@ -867,6 +867,9 @@ func (a *API) Serve(ctx context.Context) error {
 		Db:    a.mustDB(),
 		Cache: a.Cache,
 	}
+	if err := cdsService.InitMetrics(); err != nil {
+		return sdk.WithStack(err)
+	}
 	cdsService.RunTcpLogServer(ctx)
 
 	log.Info(ctx, "Starting CDS API HTTP Server on %s:%d", a.Config.HTTP.Addr, a.Config.HTTP.Port)

engine/cdn/cdn_log.go

@@ -11,6 +11,7 @@ import (
 
 	gocache "github.com/patrickmn/go-cache"
 
+	"github.com/ovh/cds/engine/api/observability"
 	"github.com/ovh/cds/engine/api/services"
 	"github.com/ovh/cds/engine/api/worker"
 	"github.com/ovh/cds/engine/api/workflow"
@@ -44,10 +45,12 @@ func (s *Service) RunTcpLogServer(ctx context.Context) {
 		for {
 			conn, err := listener.Accept()
 			if err != nil {
+				observability.Record(ctx, Errors, 1)
 				log.Error(ctx, "unable to accept connection: %v", err)
 				return
 			}
 			sdk.GoRoutine(ctx, "cdn-logServer", func(ctx context.Context) {
+				observability.Record(ctx, Hits, 1)
 				s.handleConnection(ctx, conn)
 			})
 		}
@@ -69,6 +72,7 @@ func (s *Service) handleConnection(ctx context.Context, conn net.Conn) {
 		bytes = bytes[:len(bytes)-1]
 
 		if err := s.handleLogMessage(ctx, bytes); err != nil {
+			observability.Record(ctx, Errors, 1)
 			log.Error(ctx, "cdn.log> %v", err)
 			continue
 		}
@@ -94,8 +98,10 @@ func (s *Service) handleLogMessage(ctx context.Context, messageReceived []byte)
 
 	switch {
 	case signature.Worker != nil:
+		observability.Record(ctx, WorkerLogReceived, 1)
 		return s.handleWorkerLog(ctx, signature.Worker.WorkerID, sig, m)
 	case signature.Service != nil:
+		observability.Record(ctx, ServiceLogReceived, 1)
 		return s.handleServiceLog(ctx, signature.Service.HatcheryID, signature.Service.HatcheryName, signature.Service.WorkerName, sig, m)
 	default:
 		return sdk.WithStack(sdk.ErrWrongRequest)

engine/cdn/status_handler.go

@@ -3,9 +3,23 @@ package cdn
 import (
 	"context"
 	"net/http"
+	"sync"
 
+	"go.opencensus.io/stats"
+	"go.opencensus.io/tag"
+
+	"github.com/ovh/cds/engine/api/observability"
 	"github.com/ovh/cds/engine/service"
 	"github.com/ovh/cds/sdk"
+	"github.com/ovh/cds/sdk/log"
+)
+
+var (
+	onceMetrics        sync.Once
+	Errors             *stats.Int64Measure
+	Hits               *stats.Int64Measure
+	WorkerLogReceived  *stats.Int64Measure
+	ServiceLogReceived *stats.Int64Measure
 )
 
 func (s *Service) statusHandler() service.Handler {
@@ -22,3 +36,39 @@ func (s *Service) Status(ctx context.Context) sdk.MonitoringStatus {
 	m.Lines = append(m.Lines, sdk.MonitoringStatusLine{Component: "CDN", Value: status, Status: status})
 	return m
 }
+
+func (s *Service) InitMetrics() error {
+	var err error
+	onceMetrics.Do(func() {
+		Errors = stats.Int64(
+			"cdn/tcp/router_errors",
+			"number of errors",
+			stats.UnitDimensionless)
+		Hits = stats.Int64(
+			"cdn/tcp/router_hits",
+			"number of hits",
+			stats.UnitDimensionless)
+		WorkerLogReceived = stats.Int64(
+			"cdn/tcp/worker/log/count",
+			"Number of worker log received",
+			stats.UnitDimensionless)
+		ServiceLogReceived = stats.Int64(
+			"cdn/tcp/service/log/count",
+			"Number of service log received",
+			stats.UnitDimensionless)
+
+		tagServiceType := observability.MustNewKey(observability.TagServiceType)
+		tagServiceName := observability.MustNewKey(observability.TagServiceName)
+
+		err = observability.RegisterView(
+			observability.NewViewCount("cdn/tcp/router/router_errors", Errors, []tag.Key{tagServiceType, tagServiceName}),
+			observability.NewViewCount("cdn/tcp/router/router_hits", Hits, []tag.Key{tagServiceType, tagServiceName}),
+			observability.NewViewCount("cdn/tcp/worker/log/count", WorkerLogReceived, []tag.Key{tagServiceType, tagServiceName}),
+			observability.NewViewCount("cdn/tcp/service/log/count", ServiceLogReceived, []tag.Key{tagServiceType, tagServiceName}),
+		)
+	})
+
+	log.Debug("cdn> Stats initialized")
+
+	return err
+}